In order to reduce the forming load of cold upsetting and improve the service life of mold, the optimization method of cold upsetting parameters based on escaping particle swarm algorithm was proposed, and the 3D model and the cold upsetting process flow of ball pin were introduced. Then, the Harvard floating mold was designed to solve the difficult demolding problem of ball pin. Taking reducing the forming load of cold upsetting as optimization goal, the optimized model of parameters was built. Based on Box-Behnken, twenty-seven groups of experiments were designed, and the maximum forming load was obtained by finite element software Deform-3D. Furthermore, the nonlinear relationships of optimization parameters and forming load were fit by BP neutral network, and it was verified that the fitting accuracy of BP neutral network was high. In addition, the escaping mechanism was added to particle swarm algorithm to make the algorithm have the ability to escape from local optimum, and the optimization problem of parameters was transferred into the optimal location search problem. Simulation verification results show that the maximum forming load searched by the escaping particle swarm algorithm is 11.84% less than that of the particle swarm algorithm. Finally, after production verification, the ball pin with smooth surface, full forming and no longitudinal bending meets the quality requirements of cold upsetting parts. The result indicates that through parameters searched by escaping particle swarm algorithm, the forming load is reduced, and the service life of mold is improved under the premise of meeting the quality requirement.